cheap-model-testing

Installation
SKILL.md

Cheap Model for Testing / Expensive Model for Production

Core Rule

When writing, editing, or reviewing code that calls an LLM or any pay-per-usage API (OpenAI, Anthropic, Google AI, Cohere, Mistral, Replicate, AWS Bedrock, Azure OpenAI, etc.), always default to the cheapest available model for development and testing purposes.

What To Do

During development and testing

  • Always choose the cheapest model available for the provider being used.
  • Common cheap model choices (use the latest available version):
    • OpenAI: gpt-4o-mini (or gpt-3.5-turbo if mini is unavailable)
    • Anthropic: claude-haiku-4-5-20251001 (or the latest Haiku variant)
    • Google AI / Vertex: gemini-2.0-flash (or the latest Flash variant)
    • Mistral: mistral-small-latest
    • Cohere: command-r (not command-r-plus)
    • AWS Bedrock / Azure OpenAI: whichever is the cheapest equivalent of the above
  • If the user has already specified a model and it is NOT the cheapest, proactively suggest switching to the cheapest one for testing and explain why (cost savings during development).
Related skills

More from almeidamarcell/claude-code-skills

Installs
6
GitHub Stars
4
First Seen
Feb 20, 2026